Research On Visual Tracking Technology Based On Improved Camshift Algorithm

Target tracking has been the focus and hotspot of computer vision all the time,combined with the related technologies of multidisciplinary and multi-disciplinary fields,such as automatic control,mathematical statistics and analysis,and pattern recognition.Many scholars have proposed a variety of tracking methods.However,these algorithms have more or less shortcomings,such as sensitive to light,the occlusion of the target cannot be handled,vulnerable to similar objects and so on.In practical applications,accurate tracking of targets in complex background is also a difficult point in research.In complex background,the feature of targets can easily be covered or cannot be accurately located by the interference of similar targets.In order to solve this problem,this paper carried out following works:(1)The commonly used two-dimensional color histogram model is improved,the target feature and the background feature are fused,the feature in the background which is similar to the target is reduced,the representative features of target are enlarged,a new histogram is obtained,Then the Camshift tracking can effectively handle the interference of complex background and improve the robustness of the algorithm.(2)A robot vision tracking experimental platform was set up.The industrial camera installed in the robot head to get the image and send image to the host computer,the host computer tracks the target and the target position is converted to a robot movement instruction,and then the control system drives the robot to complete the closed loop.(3)Design a position-based robot motion control algorithm.And a system coded by C + + MFC was proposed to debug.In this paper the entire system was tested in a variety of situations.When there is a similar target interference in the tracking scene,the proposed algorithm can meet the real-time requirement and can track the target accurately.When the robot is in motion control,it can stably track moving objects and accurately recognize the target under complicated environment.